Sie sind auf Seite 1von 11

Mechatronics 19 (2009) 156166

Contents lists available at ScienceDirect

Mechatronics
journal homepage: www.elsevier.com/locate/mechatronics

Design and implementation of an adaptive sliding-mode dynamic controller for wheeled mobile robots
Chih-Yang Chen, Tzuu-Hseng S. Li *, Ying-Chieh Yeh, Cha-Cheng Chang
aiRobots Laboratory, Department of Electrical Engineering, National Cheng Kung University, 1 University Road, Tainan 701, Taiwan, ROC

a r t i c l e

i n f o

a b s t r a c t
This paper designs an adaptive sliding-mode dynamic controller for wheeled mobile robots to implement the trajectory-tracking mission. First, a kinematic controller is introduced for the wheeled mobile robot. Secondly, the adaptive sliding-mode dynamic controller is proposed to make the real velocity of the wheeled mobile robot reach the desired velocity command, although the wheeled mobile robot is even with system uncertainties and disturbances. The convergence of the complete equations of motion of the wheeled mobile robot is proved by the Lyapunov stability theory. Computer simulation results illustrate the effectiveness of the proposed control schemes. Finally, the real-time experiments of the adaptive sliding-mode dynamic controller on the test ground demonstrate the feasibility of practical wheeled mobile robot maneuvers. 2008 Elsevier Ltd. All rights reserved.

Article history: Received 21 October 2007 Accepted 9 September 2008

Keywords: Adaptive control Dynamic control Kinematic control Lyapunov method Sliding-mode control Trajectory tracking Wheeled mobile robot

1. Introduction Studying the trajectory-tracking control of wheeled mobile robots (WMRs) whose motion trajectories belong to those of nonlinear systems [1] has been gaining in popularity over recent decades. Such control design has been divided into two main portions: one utilizes the kinematic trajectory tracking controller [210] to procure only the tracking issue, while the other, which is built by combining in the kinematic trajectory tracking controller and the dynamic controller of the WMR [11], is being more frequently adopted. However, it is difculty in controlling nonholonomic WMRs because, in the real world, there are uncertainties in their modeling; a few reports have even been published on control design in the presence of a system with uncertainties in the dynamic mode [1216]. The trajectory-tracking methods adopted for the kinematic control only include linearization kinematic models [4], input-output linearization [5], chained form [6,7], integrator backstepping approach [8], sliding-mode [9], and fuzzy algorithm [10]. Other researchers aim at the control architectures, the hybrid of the kinematic control, and the dynamic controller, and then propose some trajectory tracking methods. The neural network controller [11,12] is used to deal with un-modeled bounded disturbance and dynamics for the WMRs. However, the adaptability of the neural networks structure was not considered. Yang et al. [13] proposed the con-

* Corresponding author. Tel.: +886 6 275 7575x62363; fax: +886 6 234 5482. E-mail address: thsli@mail.ncku.edu.tw (Tzuu-Hseng S. Li). 0957-4158/$ - see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.mechatronics.2008.09.004

troller design of the WMR based on polar coordinates. A slidingmode control (SMC) that can overcome the external disturbance was proposed but there was an assumption that the bound of the disturbance was known. The work above, however, becomes inapplicable when the parameters of the mobile robots are not available. In order to reach a fast convergence and alleviate the control chattering effect, I/O linearization and a second order SMC technique [14] have been proposed to implement the robust output tracking control of the nonholonomic mobile robot with model uncertainties. In [15], the robust trajectory tracking problem for a mobile robot in the presence of uncertainties has been solved by means of an SMC law based on the discrete-time WMR dynamical model. No particular assumptions are needed except for the bound of the uncertainties. [1618] present a time-varying global adaptive controller at the torque level that simultaneously solves both tracking and stabilization for the WMR with unknown kinematic and dynamic parameters. This controller is based on Lyapunovs direct method and backstepping technique. Adaptive control has been applied to solving many engineering problems [1923]. The adaptive vehicle skid control [19] is designed for stability and tracking of a vehicle during slippage of its wheels without braking. Bobtsov and Nikolaev [20] illustrated a possibility for application of their adaptive algorithms to control libration angle of satellite. Wu et al. [21] realized an adaptive robust motion tracking control method for controlling an XY table driven by linear motors with a high precision. And [22] proposed an adaptive control scheme for trajectory tracking of robot manipulators in an iterative operation mode. Lim et al. [23] design a

C.-Y. Chen et al. / Mechatronics 19 (2009) 156166

157

control scheme incorporating PID control and robust model reference adaptive control to realize the levitation control of Lorentz force-type self-bearing motors. It is well known that classical SMC, which can provide great properties such as insensitivity to parameter variations and external disturbance rejection is a forceful control scheme for nonlinear systems. It has been shown in [1] that an adaptive controller can improve the closed-loop systems performance as the adaptation continues. For this reason, this study combines both the adaptive control and the sliding-mode control so as to treat both the uncertainties and the disturbances in the whole WMR system. The main contributions of this paper are (i) the velocity error between the kinematic velocity and actual velocity is adapted to set up the sliding surface; (ii) an ASMDC is proposed for the uncertain WMR system with external disturbances; (iii) all the stability analysis for the complete equations of the motion of the WMR are proven by the Lyapunov stability theory; (iv) real WMR experimental results are used to illustrate the effectiveness of the proposed control schemes. This paper is organized as follows: In Section 2, the kinematic control of the WMR is introduced. The complete equations of motion of the ASMDC based WMR is investigated in Section 3. Computer simulation results of the proposed kinematic control and ASMDC algorithms and real-time experiments of the aiRobot-W1, as shown in Fig. 1a, are given in Sections 4 and 5, respectively. Finally, Section 6 concludes the paper.

2. Kinematic controller design for WMR A typical model of a nonholonomic WMR is shown in Fig. 1b. It consists of a vehicle with two driving wheels mounted on the same axis and a free front wheel. The motion and orientation are achieved by independent actuators such as DC motors, which provide the necessary torques to the driving wheels. In the WMR model, we assume that the coordinates of the mass center of the WMR is located in the middle of the rear driving wheels. The kinematic model for the WMR under the nonholonomic constraint of pure rolling and non-slipping is given as

2 3 2 3 _ cos h 0  x  6_7 6 7 tt Jh v; y 5 4 sin h 0 5 4 xt _ 0 1 h

where J(h) 2 R32 and v 2 R2 are the full rank velocity transformation matrix and velocity vector, respectively. v(t) 2 R denotes the linear velocity of the center of mass (COM) of the WMR and x(t) 2 R denotes the angular velocity of the COM of the WMR. Eq. _ (1) is also called the steering system of WMR. The qt; qt 2 R3 are dened as

q x y hT ;

_ ___ q xyhT ;

_ _ where (x(t), y(t)) and xt; yt are the actual position and linear (translational) velocity of the COM of the WMR in the world XY coordinate, respectively. h(t) 2 R is the angle between the X-axis _ and Xc-axis and represents the heading direction of WMR, and h is the angular (rotational) velocity of the COM. Note that the mobile robot has the nonholonomic constraint, where the driving wheels roll purely and do not slip, i.e.

_ _ y cos h x sin h 0:

To formulate the tracking control problem, we dene a reference mobile robot that generates a trajectory for the actual one to follow:

_ qr Jhr tr ;

where J() is dened in (1), qr(t) = [xr (t)yr(t) hr(t)]T 2 R3 denotes the desired time-varying position and orientation trajectory, and vr(t) = [vr(t)xr(t) ]T 2 R2 denotes the reference time-varying linear and angular velocity. With regard to (3), it is assumed that the signal vr(t) is constructed to produce the desired Cartesian path qr(t). In real applications, vr(t) = [vr (t)xr(t)]T and its derivative are bounded and known. It is necessary to nd the appropriate velocity control law vc = [vc xc]T, such that q ? qr as t ? 1. In general, the trajectory tracking problem is to track a reference mobile robot with a known posture [xr(t)yr(t)hr (t)]. Therefore we dene the errors ~ between the actual and desired postures as q qr q xr x yr y hr h T . As in [4], we also dene the posture tracking error qe, which is a transformation of the reference posture qr relative to a frame xed on the actual mobile robot (Fig. 2):

32 3 xr x cos h sin h 0 6 7 6 76 7 qe 4 e2 5 4 sin h cos h 0 54 yr y 5: e3 hr h 0 0 1 e1


And the error rate becomes

3 3 2 3 2 2 3 _ 1 e2 tr cos e3 e1 6_ 7 7 6 7 6 6 7 4 e2 5 t4 0 5 x4 e1 5 4 tr sin e3 5:
Fig. 1. (a) The aiRobot-W1 and (b) simple model of a WMR.

_ e3

xr

158

C.-Y. Chen et al. / Mechatronics 19 (2009) 156166

3.1. Dynamic model of WMR The well-known dynamic equation [16] of the mobile robot system with n-generalized coordinates q 2 Rn1, and inputs r = n m, can be described as

Mqq V

_ _ m q; qq
nn

_ Fq Gq sd Bqs AT qk;

where M(q) 2 R is a symmetric positive denite inertia matrix, _ _ V m q; q 2 Rnn is the centripetal and Coriolis matrix, Fq 2 Rn1 denotes the surface friction, G(q) 2 Rn1 is the gravitational vector, sd 2 Rn1 denotes bounded unknown disturbances including unstructured unmodeled dynamics, B(q) 2 Rn(nm) is the input transformation matrix, s 2 R(nm)1 is a control input vector, A 2 Rmn is a matrix associated with the nonholonomic constraints, k 2 Rm1 is a Lagrange multiplier associated with the constraints, _ and q and q denote velocity and acceleration vectors, respectively. The variables in (8) are dened as

2
Fig. 2. Posture error coordinate.

After mentioning the posture tracking error, we have to illustrate the proposed kinematic trajectory tracking control law for the WMR. The structure of the kinematic control is depicted in dashdot line portion of Fig. 3. Eq. (7) is a common velocities command using the backstepping method, and it is recommended by many studies [8,11,16] for the tracking problem for the WMR

6 7 AT q 4 cos h 5; 0 0 I 0 2 3 cos h cos h   sr 16 7 Gq 0; Bq 4 sin h sin h 5; s ; r sl R R _ _ _ _ V m q; q 0; k mxc cos h yc sin hh;


where m is the mass of the WMR, I is the moment of inertia of the WMR about its center, and 2R and r are the distances between the two driven wheels and the radius of the wheel, respectively. The sr and sl are torque control inputs generated by the right and left DC motor, respectively. It would be more suitable to express the dynamic equations of motion in term of internal velocities. Substituting (1) and its differentiation for Eq. (8) and pre-multiplying JT(q), one can obtain

6 Mq 4 0

7 m 0 5;

sin h

tc tr cos e3 k1 e1 ; xc xr k2 tr e2 k3 tr sin e3

where k1, k2, and k3 are positive constants and tr > 0. 3. Dynamic controller design for the WMR To ensure that the motion of the WMR can follow the desired velocity generated by the kinematic controller, two dynamic controllers are introduced in this section. First, we address the dynamic model of the WMR. Second, we design a sliding-mode dynamic controller (SMDC) for the mobile robot in the torque level. Finally, the ASMDC is proposed for solving the problem raised by system uncertainties and external disturbances. The convergence of the complete equations of motion of the ASMDC based WMR is conrmed by the Lyapunov stability theorem.

_ _ _ Mqt V m q; qt Fq d Bqs; s
where M J MJ 2 R
T 22

9
m J

_ , V m J MJ V
T

2R

22

, FJ F2R

21

d ST sd , and B JT B. Since the distance between COM and the s coordinate center of the WMR is zero, the effect of V m can be eliminated from (9). The remainder variables in (9) are dened as     m 0 1 1 Mq and Bq 1 . By regarding the surface r 0 I R R friction and the disturbance torque as the modeling uncertainties and disturbances, and then the dynamic equation (9) of the simple

Fig. 3. Complete structure of ASMDC for the WMR.

C.-Y. Chen et al. / Mechatronics 19 (2009) 156166

159

model of the nonholonomic WMR, assuming all the uncertainties and disturbances are zero, is

_ tt E st;
where the system matrix E is

10

as the uncertainties of system matrix E. The ds(t) is called the external disturbance vector, which includes surface friction and disturbance torque. Suppose

  I I 1 E M 1 qBq : m r I Rm Rm

dt 11

d1 t d2 t

DE st ds t;

20

then the dynamic equation of WMR can be written as

_ tt E st dt:
3.2. SMDC design In this subsection, we utilize the SMC method to design a dynamic tracking controller which let the actual velocities of WMR converge to the control velocities generated from the kinematic controller. First, we design SMC for the nominal system, i.e. the system parameters are known and disturbances are zero. To design the torque input and estimate the desired velocities vc, we introduce the auxiliary velocity tracking error and its derivative as Hence, the SMDC (18) can be rewritten as

21

_ s seq ssw E1 tc t bec t k sgnS:

22

One should choose the switching gain k to compensate for the system uncertainties and disturbances. In order to reduce the chattering phenomenon, the most common method is to utilize the saturation function sat(S, e). Thus, replacing sgn(S) by sat(S, e) in (22) implies

ec t ec1 ec2 T tc t tt; _ _ _ ec t tc t tt:

12 13

_ s seq ssw E1 tc t bec t k satS; e;


where satsi ; e  satS; e sats1 ; e

23
sats2 ; e ,
T

Because the dynamic mode of the WMR is the rst order nonlinear function, we select the PI-type sliding surface as

 St

s1 t s2 t

 ec t b

Z
0

ec sds;

14

signsi ; jsi j > e > 0 ; i 1; 2; and e is a small positive si =e; jsi j 6 e constant. The SMDC (23) is shown in the dashed-line portion of Fig. 3. 3.3. ASMDC design It is well known that the parameter variations of the system, such as mass and inertia, are difcult to measure. And the exact value of the external disturbance is also difcult to measure in advance for practical applications. Therefore, we propose an ASMDC scheme to estimate the upper bound of jdi(t)j. Suppose that the optimal bounds of the d* and c* exist. The adaptive algorithm for the bound of c* is

where b is the sliding-surface integral parameter and b > 0. It is noted in (14) that once the system is on the sliding surface S = 0, Rt then ec t b 0 ec sds and it is obvious that the tracking error ec (1) ? 0 as b > 0. Hence it is desired to drive the closed-loop system toward the sliding surface S = 0. Meanwhile, with the derivative _ S of sliding surface S(t), one can obtain that

_ _ St ec t bec t:

15

It obviously that the tracking error ec(1) ? 0 if the integral parameter, b, is selected properly. From the concept of the equivalent control, the equivalent control law seq is obtained by recogniz_ ing that St is equal to zero is a necessary condition for the state trajectory to stay on the sliding surface. Thus, substituting (10) for (15), one can obtain

"

_ ^t c

_ ^ c1 0

0 _ ^ c

q1 s1 sats1 ; e
0

 0 ; q2 s2 sats2 ; e

24

_ where ^t is the estimated value of c* and is also used as the SMC c switching gain later. The qi (i = 1,2) is denoted as a positive adaptation gain. The ASMDC for the WMR is designed as follows:

_ _ St tc t E st bec t 0:
Therefore, the equivalent control law seq is

16

_ ^ s E1 tc t bec t ct satS; e:
And we dene the estimation error as

25

_ seq t E1 tc t bec t;
where E
1

17

~t ^t c : c c

26

is selected as

E1

  r Rm I ; 2R Rm I

The design task is to choose an adaptive law to update the estimate ^t such that S(t) converges to a zero vector, limt?1S(t) = 0. c The complete structure of ASMDC for the WMR is shown in Fig. 3. Theorem 1. Both the posture tracking error qe and the velocity tracking error vc of the complete equations of motion of the WMR with uncertainties and disturbances (21) will asymptotically converge to zero vectors, if the kinematic controller (7) and the ASMDC (25) are applied. Proof. Let the Lyapunov function candidate be dened as

and the equivalent control law seq can make the system state remain on the sliding surface if the dynamic is known exactly. However, if there are system uncertainties in the real application, in order to satisfy the sliding condition, we must employ the discontinuous control law ssw in (17). Thus the control law is composed of equivalent control seq and switching control ssw, i.e.

_ s seq ssw E1 tc t bec t k sgnS;


where k1 k 0   0 , k2 and
T

18
constant, and

L L1 L2 ;
where

27

ki

is

positive

sgnS sgns1 sgns2 . Then the dynamic Eq. (10) in the presence of parameter uncertainties and external disturbances becomes

_ tt E st ds t E st DE st ds t;

19

1 2 1 cos e3 ; e e2 2 2 1 k2 1 1 ~ ~1 ~2 L2 St; ct ST tSt 1=q1 c2 t 1=q2 c2 t: 2 2 L1 e1 ; e2 ; e3

28 29

where E is denoted as the nominal part of system matrix E introduced by WMR parameters r,R,m and inertia I. The DE is denoted

Clearly, L P 0. Substituting (6) and (7) for the time derivative of L1 in (28), we obtain

160

C.-Y. Chen et al. / Mechatronics 19 (2009) 156166


2

k3 tr sin e3 _ L1 k1 e2 6 0: 1 k2

30

step 4: step 5: step 6: step 7: step 8:

Create the dynamic equation from (21). Drive the complete control law with ssw (23). _ ^ Decide the adaptive law ct from (24). Calculate the overall control s from (25). Adjust the parameters, k1,k2, and necessary. k3, given in (7) if

_ Therefore, if the reference velocity vr P 0 then L1 6 0. Differentiating (29), one can obtain

_ _ _ _ ~ ~ ~ ~ L2 St; ~t ST t St 1=q1 c1 c1 1=q2 c2 c2 c _ _ _ ~ ~ S t tc t tt bec t 1=q1 c1 c1


T

_ ~ ~ 1=q2 c2 c2 :
(25) for (31) implies

31

_ _ ^ ~ ct is equal to ct because the c* is constant. Substituting (21) and 4. Simulation results Computer simulation results are given to demonstrate the effectiveness of all the proposed control schemes. To observe and compare the simulation results more easily, we choose two kinds of reference trajectories for the simulations: one is a straight line trajectory, and the other is a round rectangle one. System parameters of the WMR are shown in Table 1.

_ _ ~ ^ ~ ^ L2 St; ct ST t ct satS; e dt 1=q1 c1 c1 _ ~ ^ 1=q2 c2 c2 ~ ST t c ct satS; e dt 1=q1 _ _ ~ ^ ~ ^ c1 c1 1=q2 c2 c2 ST t c satS; e dt _ ^ 1=qi ci si satsi ; e:
From adaptive law (24), one can get
2 X i1

~ ci t 32

4.1. The straight line trajectory The straight line trajectory was generated from the desired velocity vr(t) = 0.5(m/s) and the angular velocity is zero. The initial posture of the reference trajectory is set at qr(0) = [xr(0)yr(0)hr(0)]T = [0045]T. The actual initial posture of the WMR is q(0) = [x(0)y(0)h (0)]T = [31180]T. (1) Nominal Condition: System parameters of the WMR are shown in Table 1. In this case, there are no uncertainties and disturbances. Evidently, both of these two architectures have a nice tracking ability, as shown in Fig. 4, and this also proves that no matter what architectures we use, these two controllers can trace the velocity command generated by the kinematic controller. (2) Real Parameter Variations and Disturbance Condition: At 15 s, the WMRs actual mass and inertia increase 50% and 100%, respectively. That is,

_ ^ 1=qi ci si satsi ; e 0; t P 0:
Substituting (33) for (32) implies

33

_ ~ L2 St; ct ST t c satS; e dt c s1 sats1 ; e c s2 sats2 ; e d1 ts1 t 1 2 d2 ts2 t 6


2 X jdi tj si satsi ; e c si satsi ; e i i1 2 X si satsi ; ec jdi tj 6 0: i i1

34

_ _ ^ The adaptive law ct ensures L2 is negative semi-denite. With the ~ adaptive law (24), L2 St; ct as a function of t does not increase, that is

~ ~ L2 St; ct 6 L2 S0; c0; 8t P 0:

m : 4 ! 6 kg I : 2:5 ! 5 kg m2

35

t P 15 s:

~ Therefore, the S(t) and ct are bounded. The implementation of adaptive law (24) is

^ ci t qi

Z
0

^ si satsi ; eds ci 0;

36

^ where ci 0 is an initial estimate of the upper bound of c . From (30) i _ and (34) we can conclude that L is negative semi-denite. That is, the posture tracking error qe and the sliding surface S approach zero Rt vectors. It is noted in (14) that once S = 0, then ec t b 0 ec sds and it is obvious that ec(1) ? 0. This completes the proof of the theorem. h Remark 1. Based on the kinematic controller (7) and the ASMDC (25), one can spur the WMR (8) to keep up with the desired trajectory (4). And the design procedure for ASMDC can be described as follows: step 1: step 2: step 3: Determine a stable sliding surface S from (14). Drive the equivalent control seq from (17). Separate the nominal part E from the system matrix E as (19).

The assumption for the uncertainty in the robots parameters, as revealed in Fig. 5, particularly the mass and hence the inertia can be justied in real applications such as in the automotive manufacturing industry and warehouse, where the WMR carries a variety of parts of different shapes and masses. It has slight effect on the SMDC, but the effect on ASMDC is almost zero. But at 25 s, the external disturbances are fed in, where a disturbance is introduced with a magnitude bounded by 6, that is, jds(t 25)j = j randon(t 25)j 6 6. Obviously, the efcacy of the SMDC has dropped drastically, but the efcacy of the ASMDC is still maintained.

Table 1 Actual parameters of the wheeled mobile robot Parameter m I R r Description Mass of the WMR Inertia of the WMR about its center Distance between the two driven wheels Radius of the wheels Nominal value 4 kg 2.5 kg m2 0.15 m 0.03 m

C.-Y. Chen et al. / Mechatronics 19 (2009) 156166

161

Fig. 4. The comparisons of the tracking ability between the SMDC and the ASMDC with combining the kinematic controller. (a) The real trajectories of these two controllers. (b) The tracking errors in X-coordinate. (c) The tracking errors in Y-coordinate. (d) The tracking errors in the directional angles. (e) Linear velocities. (f) Angular velocities. (g) The variations in the update law. (h) The control signal.

162

C.-Y. Chen et al. / Mechatronics 19 (2009) 156166

Fig. 5. The comparisons of the tracking ability between the SMDC and the ASMDC with combining the kinematic controller under parameter variations and disturbances. (a) The real trajectories of these two controllers. (b) The tracking errors in X-coordinate. (c) The tracking errors in Y-coordinate. (d) The tracking errors in the directional angles. (e) Linear velocities. (f) Angular velocities. (g) The variations in the update law. (h) The control signal.

4.2. The round rectangle trajectory In this simulation, the round rectangle trajectory was generated by switching the desired velocity and the angular velocity in time,

tr

0:785 m=s; 0 10 n s < t 6 3:146 10n s

xr

1 m=s; t > 3:146 10n s  0:5 rad=s; 0 10n s < t 6 3:146 10n s 0; t > 3:146 10n s

C.-Y. Chen et al. / Mechatronics 19 (2009) 156166

163

where n, which changes from 0 to 3 step by step, is a positive constant. And the initial posture of the reference trajectory is

qr(0) = [xr(0)yr(0)hr(0)]T = [01.6270]T. The actual initial posture of the WMR is q(0) = [x(0)y(0)h (0)]T = [2 1180]T.

Fig. 6. The comparisons of the tracking ability between the SMDC and the ASMDC with combining the kinematic controller. (a) The real trajectories of these two controllers. (b) The tracking errors in X-coordinate. (c) The tracking errors in Y-coordinate. (d) The tracking errors in the directional angles. (e) Linear velocities. (f) Angular velocities. (g) The variations in the update law. (h) The control signal.

164

C.-Y. Chen et al. / Mechatronics 19 (2009) 156166

(1) Nominal condition: To simplify the complexity of the simulation, the system parameters of the WMR are all the same as in the previous simulation. In this case, there are no uncertainties and disturbances. In this situation, both of these

two architectures have a similar tracking ability but in the corners, as shown in Fig. 6f, the SMDC method has bigger angle errors than the ASMDC.

Fig. 7. The comparisons of the tracking ability between the SMDC and the ASMDC with combining the kinematic controller under parameter variations and disturbances. (a) The real trajectories of these two controllers. (b) The tracking errors in X-coordinate. (c) The tracking errors in Y-coordinate. (d) The tracking errors in the directional angles. (e) Linear velocities. (f) Angular velocities. (g) The variations in the update law. (h) The control signal.

C.-Y. Chen et al. / Mechatronics 19 (2009) 156166

165

Fig. 8. The real tracking ability on the aiRobot-W1. (a) Overview of experimental setup. (b) The kinematic controller. (c) Combining the SMDC with kinematic controller. (d) Combining the ASMDC with kinematic controller.

(2) Real parameter variations and disturbance condition: The parameter variations, generated at 15 s, are the same as in the preceding simulation. And the external disturbances, which are bounded at 8, are injected into the system at 25 s. The noise inuences on the SMDC are much bigger than those on the ASMDC. In Fig. 7e and f, the linear velocity and angular velocity of the ASMDC could keep up with the desired ones more quickly than in the SMDC. Fig. 7g presents the adaptive law (36) and they tend toward some nite values. In the simulation result, we illustrate that the ASMDC is robust enough to resist the parameter variations and the external disturbances. Remark 2. From the simulation results, one can easily nd that the ASMDC has much better tracking performances than the SMDC does, such as its ability to deal with the uncertainties and the external disturbances. Moreover, the ASMDC can also reduce the tracking errors at each turn in the round rectangular trajectory tracking, where the desired linear velocity and the desired angular velocity switch abruptly.

5. Experiments In this section, we want to use the developed aiRobot-W1 to realize kinematic controller, SMDC with kinematic controller, and ASMDC with kinematic controller separately. The experimental setup is shown in Fig. 8a and is established by integrating the real-time image processing module, wireless communication module, and strategy decision module. The specication of the host computer adopted here is the Intel Pentium 4 CPU 3.00 G Hz with 2G SRAM. The equipment for the experiments is listed as follows: the DCR-PC330 camera, manufactured by SONY, is used to cap-

ture the image in the test eld. The Meteor, manufactured by Matrox Electronic Systems Ltd, is adopted to grab real-time image data and transmit them to an external interface or displayed RAM. The SST-2400EXT wireless radio modem was developed by ICP DAS as the communication media between the host computer and the WMRs, where the transmitter of the wireless communication system connects with the host computer and the receiver mounted on the aiRobot-W1. The low level controller, TI TMS320F240 DSP, which is set up on the aiRobot-W1, takes charge of the motion control. The aiRobot-W1 consists of the low level controller, DC motors, receiver, structural mechanism, and electric circuits. In this experiment shown in Fig. 8, the desired trajectory was chosen as a round rectangle and is drawn with a red line. It can be used to verify both the straight and the curve tracking performance at the same time. The dimensions of the experimental eld are 3 m in length and 2.5 m in width. Each of the actual experimental photographs is taken from 25 to 30 picture frames in a video le captured by a charge-coupled device (CCD) camera, and each video le is approximately 2.5 min long. Fig. 8b presents the tracking ability of the kinematic controller. It is evident that the tracking errors at the corners are considerable. The tracking ability of the SMDC with kinematic controller is revealed in Fig. 8(c), where the trajectory is better than the one generated by the kinematic controller. However, the ASMDC with kinematic controller scheme is observed to have the best tracking ability with fewer tracking errors, as indicated in Fig. 8d. 6. Conclusions In this paper, we have proposed a two-stage tracking controller, combined the kinematic controller and ASMDC, for a WMR with unknown parameter variations and external disturbances, which

166

C.-Y. Chen et al. / Mechatronics 19 (2009) 156166 [6] Jiang ZP, Nijmeijev H. A recursive technique for tracking control of nonholonomic system in chained form. IEEE Trans Automat Control 1999;44(2):26579. [7] Murray RM, Sastry S. Nonholonomic motion planning: steering using sinusoids. IEEE Trans Automat Control 1993;38(5):70016. [8] Jiang ZP, Nijmeijev H. Tracking control of mobile robots: a case study in backstepping. Automatica 1997;33(7):13939. [9] Bloch AM, Drakunov S. Stabilization and tracking in the nonholonomic integrator via sliding modes. Syst Control Lett 1996;29(2):919. [10] Horacio MA, Simon GG. Mobile robot path planning and tracking using simulated annealing and fuzzy logic control. Expert Syst Applicat 1998;15:4219. [11] Fierro R, Lewis FL. Control of a nonholonomic mobile robot: backstepping kinematics into dynamics. J Robot Syst 1997;14(3):14963. [12] Fierro R, Lewis FL. Control of a nonholonomic mobile robot using neural networks. IEEE Trans Neural Networks 1998;9(4):589600. [13] Yang JM, Kim JH. Sliding mode control for trajectory tracking of nonholonomic wheeled mobile robots. IEEE Trans Robot Automat 1999;15(3):57887. [14] Hu Z, Li Z, Bicker R, Marshall C. Robust output tracking control of nonholonomic mobile robots via higher order sliding mode. Nonlinear Stud 2004;11(1):2335. [15] Corradini ML, Orlando G. Control of mobile robots with uncertainties in the dynamic model: a discrete time sliding mode approach with experimental results. Control Eng Prac 2002;10(1):2334. [16] Kukao T, Nakagawa H, Adachi N. Adaptive tracking control of nonholonomic mobile robot. IEEE Trans Robot Automat 2000;16(5):60915. [17] Do KD, Jiang ZP, Pan J. Simultaneous tracking and stabilization of mobile robots: an adaptive approach. IEEE Trans Automat Control 2004;49(7): 114752. [18] Kim MS, Shin JH, Hong SG, Lee JJ. Designing a robust adaptive dynamic controller for nonholonomic mobile robots under modeling uncertainty and disturbances. Mechatronics 2003;13(5):50719. [19] Kececi EF, Gang T. Adaptive vehicle skid control. Mechatronics 2006;16(5): 291301. [20] Bobtsov A, Nikolaev N, Slita O. Adaptive control of libration angle of a satellite. Mechatronics 2007;17(4-5):2716. [21] Wu J, Pu D, Ding H. Adaptive robust motion control of SISO nonlinear systems with implementation on linear motors. Mechatronics 2007;17(45):26370. [22] Ouyang PR, Zhang WJ, Gupta MM. An adaptive switching learning control method for trajectory tracking of robot manipulators. Mechatronics 2006;16(1):5161. [23] Lim T M, Zhang D. Control of Lorentz force-type self-bearing motors with hybrid PID and robust model reference adaptive control scheme. Mechatronics 2008;18(1):3545.

allows the WMR to achieve complete tracking of the desired trajectory. The kinematic controller, a velocity controller for the kinematic system, is introduced rst. And then we employ a direct adaptive algorithm, which is designed to compensate for the upper bound of system uncertainties and external disturbances and to design SMDC for navigating the WMR. The ASMDC is designed to make the real mobile robot velocity reach the desired velocity command determined by the kinematic controller. Surveying the existed literature about WMRs, we found that this paper is the rst one adopting the velocity error between the kinematic velocity and actual velocity to set up the PI-type sliding surface. Both the posture tracking error qe and the velocity tracking error ec of the complete equations of motion of the WMR with uncertainties and disturbances have been shown to converge to zero vectors asymptotically. All the computer simulations and practical experimental results demonstrate that the proposed scheme is indeed feasible and effective. Acknowledgement The support of this work in part by the National Science Council of the Republic of China under NSC 95-2221-E-006-363-MY2 is gratefully acknowledged. References
[1] Slotine JJE, Li W. Applied nonlinear control. Imprint Englewood Cliffs, NJ: Prentice Hall; 1991. [2] Wit CC, Sordalen OJ. Exponential stabilization of mobile robots with nonholonomic constraints. IEEE Trans Automat Control 1992;13(11):17917. [3] Sun S. Designing approach on trajectory-tracking control of mobile robot. Robot Comput-Integrated Manufact 2005;21(1):815. [4] Kanayama Y, Kimura Y, Miyazaki F, Noguchi T. A stable tracking control method for an autonomous mobile robot. In: Proceedings of the IEEE conference robotics and automation; 1990. p. 3849. [5] Kim DH, Oh TH. Tracking control of a two-wheeled mobile robot using inputoutput linearization. Control Engineer Prac 1999;7:36973.

Das könnte Ihnen auch gefallen